Mehmood Sheikh
AI Engineer specializing in autonomous agents, RAG architectures, and real-time voice pipelines.
Based in: Karachi, PakistanCurrently: AI/ML Engineer, ServeTech GlobalExperience: 2 years
Mehmood Sheikh is an experienced AI Engineer with a proven track record of building production-grade systems using Python, FastAPI, and vector databases. He focuses on developing scalable, product-first AI solutions that eliminate business bottlenecks and operational costs through advanced automation and voice-based agents.
Experience
AI/ML Engineer
ServeTech GlobalOnsiteworkMay 2026 – Present
Summary
Faced with missed client calls at PalmLeaf Massage & Wellness, I engineered the Grace AI Receptionist, a 24/7 voice and chat agent powered by n8n, Vapi, and FastAPI. I architected a background-syncing system that delivers instant 2.5-second responses. To eliminate the need for human callers, I architected a highly scalable, serverless AI Outbound Sales Agent (Python, Telnyx, Google Workspace) that autonomously qualifies leads, handles objections, and books meetings. Managed entirely via a single Google Sheet, the system dynamically scales to process 200+ calls daily for just ~$235/month, allowing the team to instantly adjust concurrent call volume with zero server maintenance. To deliver instant real-time risk assessments in Ready2Go, I built a centralized data engine that automatically ingests live feeds from 11 different government APIs, completely eliminating slow dashboard loading times. To automate the review of complex business continuity plans in Ready2Go, I engineered an end-to-end AI pipeline utilizing a Weaviate vector database to read, process, and evaluate user-uploaded documents to assign automated quality scores. Faced with AI hallucination issues in risk reporting, I designed strict prompt guardrails that force the model to extract only hard facts like exact dates and financial damages, stopping the AI from guessing.
What I did
- Structured deployments by isolating FastAPI backend, PostgreSQL, Weaviate, and n8n background workers into separate Docker containers.
- Implemented asynchronous FastAPI backend to handle concurrent voice streams without blocking the main thread.
- Scaled backend services using Gunicorn with multiple Uvicorn workers behind a load balancer.
- Built an agentic workflow for the Business Continuity Vault project to read and score large corporate risk documents.
Results
- Delivered instant 2.5-second responses for voice/chat agent.
- Scaled outbound sales system to 200+ calls daily for ~$235/month.
- Eliminated slow dashboard loading times by ingesting 11 government APIs.
- Automated quality scoring for business continuity plans using Weaviate.
- Eliminated AI hallucinations in risk reporting through strict prompt guardrails.
Pythonn8nVapiFastAPITelnyxGoogle WorkspaceGoogle SheetsWeaviateVector DatabaseAPIsGunicornUvicornllama.cppOllamaGGUFQuantizationUbuntuDocker
A.I. Specialist
Intellexal Solutions Private LimitedOnsiteworkJul 2025 – Apr 2026
Summary
Designed the Lumina core system connecting a standard phone network to an advanced conversational AI, enabling the company to deploy intelligent, voice-based agents over phone calls. Built an N8N-style Agentic AI System in Lumina that connects chatbots to 20+ business tools (Google Sheets, CRMs, APIs) allows AI to perform actions like saving data to Google Sheets, checking order status, booking appointments. To enable sophisticated, human-like automated conversations, I engineered an end-to-end voice AI pipeline (VAD → STT → LLM → TTS) with dynamic persona templates, empowering the agent to conduct nuanced, role-specific dialogues. Enabled the AI to provide highly accurate, client-specific answers by building a custom knowledge base for each client using a Weaviate vector database and an information retrieval (RAG) pipeline. To create bilingual (Urdu/English) voice for the AI, Currently fine-tuning a Text-to-Speech (TTS) model on Common Voice dataset, eliminating the robotic sound for a more natural user experience.
What I did
- Built an async FastAPI backend with PostgreSQL logging to manage concurrent call volume.
- Used VAD to filter background noise and paired transcripts with a Weaviate RAG pipeline to maintain accuracy on noisy phone lines.
Results
- Connected chatbots to 20+ business tools.
- Engineered end-to-end voice AI pipeline (VAD, STT, LLM, TTS).
- Implemented RAG pipeline with Weaviate for client-specific knowledge bases.
- Fine-tuned TTS model on Common Voice dataset for bilingual support.
- Maintained end-to-end response times between 1.5 and 2.5 seconds for the full VAD, STT, LLM, and TTS loop.
- Handled around 50 simultaneous calls without audio drops or database bottlenecks.
n8nGoogle SheetsCRMsAPIsVADSTTLLMTTSWeaviateRAGCommon Voice
AI/ML Developer
FameGuildRemotecontractNov 2024 – Jun 2025
Summary
Faced with slow marketing decision-making, I designed an AI agent to generate marketing strategies, forecast trends, and answer key business questions - reducing planning time by 40% and improving campaign accuracy. To boost brand visibility across platforms, I built an automation tool that created and scheduled social media posts with personalized captions & dynamic images, smart hashtags, and ideal timing - leading to a 60% increase in engagement and saving 30+ hours per week. To support NFT model training, I developed a pipeline that automatically collected and cleaned NFT images and metadata - accelerating dataset preparation by 5× and ensuring seamless input for text-to-image AI models.
Results
- Reduced planning time by 40% and improved campaign accuracy.
- Increased engagement by 60% and saved 30+ hours per week through automation.
- Accelerated NFT dataset preparation by 5×.
AI AgentsText-to-image AIAutomation Tools
Skills
Technical
PythonPandas
Weaviate
n8n
Fast APIVapi
LangChain
NumPy
SQL
PostgresSQLTelnyx
Deepgram
ElevenLabs
PyTorchscikit-learn
Hugging Face
Transformers
LangGraph
Eliza
Matplotlib
Seaborn
Docker
MongoDB
TensorFlowKeras
OpenClaw
Real-time state synchronization
Quantization
Gunicorn
Uvicorn
Asynchronous Programming
llama.cppOllama
Structured Decoding
Pydantic
Interruption handling
Languages
Urdu
English
Projects
Rahbar - UpliftAI Hackathon
Lead Developer / Designerpersonal
To make the web accessible to millions in Pakistan facing literacy barriers, I designed and built 'Rahbar,' a conversational AI that transforms the visual web into a fully interactive audio experience, enabling users to browse, search, and understand content using only their voice. Engineered an intelligent page analysis agent that identifies all interactive elements (links, forms, buttons) and provides step-by-step vocal guidance, effectively translating complex visual layouts into simple, actionable audio commands (e.g., 'Say click contact us to proceed'). To expand Rahbar into a full productivity partner, I integrated Google Suite APIs (Gmail, Calendar) and local file system operations, empowering users to manage emails, check schedules, and perform file CRUD tasks purely through conversation.
Conversational AIGoogle Suite APIsGmail APICalendar APIFile System Operations
Self-correcting RAG and structured-order workflow
personal
Built a cyclic graph designed to prevent AI hallucinations using a strict TypedDict state schema to track conversation history, extracted variables, and retry counts.
LangGraphPostgreSQLPostgresSaverPydanticLangChain
Local Multi-Agent RAG Prototyping Environment
personal
Developed a local development environment using a dual-GPU setup (RTX 3090 and RTX 40-series) to host and optimize models using llama.cpp and Ollama. Utilized GGUF formats and 4-bit/8-bit quantization to prototype multi-agent RAG pipelines and test structured outputs with real-time token streaming.
llama.cppOllamaGGUF4-bit quantization8-bit quantizationUbuntu
Business Continuity Vault
personal
Built an agentic workflow to read and score large corporate risk documents, using a local model for heavy text processing and a cloud API for final structured data output.
OllamaWeaviateGPT-4o
Education
XSEED Incubation Program 2.0
INSEADEntrepreneurshipOct 2025
Societies & activities
XSEED Incubation Program
Bachelor of Science in Software Engineering
NED University of Engineering & TechnologySoftware EngineeringFeb 2021 – Feb 2025
Societies & activities
Usman Institute of Technology (Affiliated with NED)
Recognition
Awards
IELTS ACADEMIC - CEFR Level C1
IELTS
Overall Band Score: 7.0, TRF No.: 25PK502450SHEM363A
Certifications
Fundamentals of AI Agents Using RAG and LangChain
Coursera
Mathematics for Machine Learning: Linear Algebra
Coursera
Full-Stack Deep Learning with Python
LinkedIn Learning
PyTorch Essential Training: Deep Learning
LinkedIn Learning
Recurrent Neural Networks
LinkedIn Learning
Python Data Structures
Coursera
Programming for Everybody (Getting Started with Python)
Coursera
Mehmood's twin is AI, it can make mistakes.
More people on Kanvis
Muhammad Umer
Lahore
AI Engineer specializing in multi-step LLM systems and document intelligence pipelines.
n8n · LLM Extraction · Python · Slack API (Bolt)
kanvis.me/muhammad-umer1
Haris Majeed
Lahore
AI Engineer specializing in Computer Vision pipelines and Multimodal Deepfake Detection.
Python · RF-DETR · scikit-learn · Git
kanvis.me/haris-majeed
Fatima Bilal Hassan
Lahore
CS at LUMS | AI & RAG Specialist | Full-Stack Developer building for scale and social impact
NLP · RAG pipelines · LangChain · Python
kanvis.me/fatima-bilal-hassan
Shahid Shafiq
Karachi
Creative Head | Digital Strategist | 17+ Years of Design, AI & Growth Innovation
Hindi · Urdu · Digital Marketing · English
kanvis.me/shahid-shafiq
